Assignment5.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h1 style=\"color:rgb(0,120,170)\">Assignment 5: CharRNN (using PyTorch)</h1>\n",
    "Build an character level LSTM using pytorch but without using the built in LSTM.\n",
    "\n",
    "Test and train-set: https://github.com/ryanmcdermott/trump-speeches"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# %load_ext line_profiler # for prun/lprun\n",
    "%matplotlib inline\n",
    "import re\n",
    "import os\n",
    "from pathlib import Path, PosixPath\n",
    "from typing import Tuple\n",
    "from collections import defaultdict\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "from tqdm import tqdm\n",
    "from torch import tensor\n",
    "from torch.utils.data import Dataset, DataLoader\n",
    "from torch.distributions.multinomial import Multinomial\n",
    "from functools import partial\n",
    "\n",
    "# use gpu if you can\n",
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2 style=\"color:rgb(0,120,170)\">Dataloader and preprocessing</h2>\n",
    "\n",
    "As most chars are irrelevant and the training set includes multiple uncommon unicode signs (often just once), this blows up the network size. To simplify training, do some character-level-preprocessing, so that the one-hot encoded representation is smaller."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "REPLACE = {\"”\": '\"', \"“\":'\"', \"$\": \"\", \"—\": \"-\", \"–\": \"-\", \"é\":\"e\", \"…\": \"...\", \"‘\": '\"', \"’\": \"'\", \n",
    "           \"%\": \"percent\", \"@\": \"at\", \";\": \",\", \"/\": \" or \", \"_\": \"\", \"=\": \"\"}\n",
    "\n",
    "class Transcoder():\n",
    "    \"\"\" The TextDS shouldn't contain encoders, as it is transferred to multiple dataloader processes \"\"\"\n",
    "    def __init__(self, text_file: PosixPath):\n",
    "        data = open(text_file, encoding=\"utf-8\").read()\n",
    "        data = self.preprocess(data)  # encoding on windows isn't utf-8\n",
    "        self.decoder = sorted(set(data))\n",
    "        encoder = {s: i for i, s in enumerate(self.decoder)}\n",
    "        # if sign wasn't in training set, replace it with \" \" in encode\n",
    "        self.encoder = defaultdict(lambda: encoder[\" \"], encoder)\n",
    "        print(f\"Created en/de-coder for {text_file} with {len(self.decoder)} letters\")\n",
    "\n",
    "    def decode(self, data: list) -> str:\n",
    "        return \"\".join([self.decoder[d] for d in data])\n",
    "    \n",
    "    def encode(self, data: str) -> np.ndarray:  # slicing on np arrays is easier \n",
    "        return np.array([self.encoder[d] for d in data], dtype=np.byte)\n",
    "        \n",
    "    def preprocess(self, data: str) -> str:\n",
    "        data = data.lower()  # 50% less letters -> 50% less computing. and it's bad in the original anyways\n",
    "        data = re.sub(\"\\(.+\\)\", \"\", data)  # i.e. (CHEERS)\n",
    "        data = re.sub(\"\\[.+\\]\", \"\", data)  # i.e. [SILENCE]\n",
    "        data = re.sub(r\"\\d\\d?/\\d\\d?/\\d\\d\\d\\d\", \"\", data)  # i.e. 7/11/2016, has to be before letter-replacer\n",
    "        data = re.sub(r\"q&a\", \"\", data)  # trash that appears once and adds an &\n",
    "        data = re.sub(r\"speech \\d\\n\", \"\", data)  # i.e. SPEECH 1\n",
    "        # replace letters with static replace-map, normally we'd use unicodedata.normalize\n",
    "        data = \"\".join([REPLACE[letter] if letter in REPLACE else letter for letter in data])\n",
    "        data = re.sub(r\"\\n+\", \"\\n\", data)\n",
    "        # debug option: show all remaining chars and their charcodes\n",
    "        # print({ord(d): d for d in sorted(set(data)-set(\"abcdefghijklmnopqrstuvwxyz0123456789\"))})\n",
    "        return data.strip()\n",
    "        \n",
    "\n",
    "class TextDS(Dataset):\n",
    "    def __init__(self, text_file: PosixPath, seq_length: int, transcoder: Transcoder, use_wordstarts=True):\n",
    "        data = transcoder.preprocess(open(text_file, encoding=\"utf-8\").read())  # encoding on windows isn't utf-8\n",
    "        self.seq_length = seq_length\n",
    "        self.data = transcoder.encode(data)\n",
    "        if use_wordstarts:\n",
    "            self.wordstarts = [match.start()+1 for match in re.finditer(\"[ \\n]+\", data)]\n",
    "            self.wordstarts = [0] + list(filter(lambda a: a < len(data) - seq_length, self.wordstarts))\n",
    "        else: # yields more samples but longer training time\n",
    "            self.wordstarts = range(len(data) - seq_length)\n",
    "        print(f\"Created dataloader for {text_file} with {len(self.wordstarts)} samples.\")\n",
    "        # load all tensors into gpu once\n",
    "        self.items = [(tensor(self.data[self.wordstarts[idx]:self.wordstarts[idx]+self.seq_length], dtype=torch.int64).to(device), \n",
    "                tensor(self.data[  self.wordstarts[idx]+self.seq_length], dtype=torch.int64).to(device))\n",
    "                      for idx in range(len(self.wordstarts))]\n",
    "        \n",
    "    def __len__(self) -> int:  # all possible slices\n",
    "        return len(self.wordstarts)\n",
    "        \n",
    "    def __getitem__(self, idx: int) -> Tuple[torch.Tensor, torch.Tensor]:\n",
    "        return self.items[idx]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2 style=\"color:rgb(0,120,170)\">The pytorch model</h2>\n",
    "\n",
    "Unlike numpy, we don't need a backward pass, as pytorch can automatically do this with autograd, so we only need the forward-formulas from the last exercise.\n",
    "\n",
    "This version of the model also has variables for the initialization (init0) and activation functions, as various versions of LSTMs can have different variants."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Model(nn.Module):\n",
    "    def __init__(self, n_cells: int, n_output: int, **kwargs):\n",
    "        super(Model, self).__init__()\n",
    "        # we don't need n_input, as torch is cool enough to handle variable length sequences just fine\n",
    "        self.n_cells = n_cells   # hidden size\n",
    "        self.n_output = n_output # alphabet size\n",
    "        \n",
    "        # input connections\n",
    "        self.i, self.o, self.z = [nn.Linear(self.n_output, self.n_cells) for _ in range(3)]\n",
    "        # recurrent connections, without bias as we allready have the additive bias from input connections\n",
    "        self.ri, self.ro, self.rz = [nn.Linear(self.n_cells, self.n_cells, bias=False) for _ in range(3)]\n",
    "        # output connection\n",
    "        self.v = nn.Linear(self.n_cells, self.n_output)\n",
    "        # activation functions\n",
    "        self.a0, self.a1, self.a2 = [nn.Sigmoid(), nn.Tanh(), nn.Softmax((1))]  # softmax on dim 2\n",
    "        # previous initializations\n",
    "        self.init0 = torch.zeros(self.n_cells).to(device)\n",
    "        \n",
    "        \n",
    "    def forward(self, inp: torch.Tensor, **kwargs) -> torch.Tensor:\n",
    "        batch_size, n_input = inp.shape\n",
    "        oh = nn.functional.one_hot(inp, self.n_output).float()\n",
    "        \n",
    "        # init recurrent stuff with 0\n",
    "        i = o = z = c = self.init0\n",
    "        # actual LSTM logic\n",
    "        for index in range(n_input):\n",
    "            t = oh[:,index,:]\n",
    "            i = self.a0(self.i(t) + self.ri(i))\n",
    "            o = self.a0(self.o(t) + self.ro(o))\n",
    "            z = self.a1(self.o(t) + self.rz(z))\n",
    "            c = c + i * z\n",
    "            h = o * self.a1(c)\n",
    "        return self.a2(self.v(h))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2 style=\"color:rgb(0,120,170)\">Training and evaluation loop</h2>\n",
    "\n",
    "To train the model above, we need a classic data loading setup as well as a training loop. This can also be used for hyperparameter search, and the most popular hyperparameters like  epochs, word_len, batch_size, number of lstm cells, learning rate....) are already parameters to ease hyperparameter search. However, most parameters here are rather big, and further hyperparameter search could increase the training times by orders of magnitude.\n",
    "\n",
    "Different kinds of data and setups may additionally require different optimizers, encodings, activation functions...\n",
    "\n",
    "Due to our dataloader setup, we even have a reduced and an extended dataset. To speed up learning, we do pretraining with the reduced set and then do a fine tuning iteration with a smaller learning rate.\n",
    "\n",
    "Progress is visualized with the tqdm module, history is tracked in a [2, epochs] np array."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Created en/de-coder for trump_train.txt with 46 letters\n",
      "Created dataloader for trump_train.txt with 163327 samples.\n",
      "Created dataloader for trump_val.txt with 4166 samples.\n",
      "preprocessing done\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|███████████████████████████████████████████████████████| 100/100 [30:45<00:00, 18.45s/ep, tloss=6.66, vloss=0.243]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Created en/de-coder for trump_train.txt with 46 letters\n",
      "Created dataloader for trump_train.txt with 867175 samples.\n",
      "Created dataloader for trump_val.txt with 23629 samples.\n",
      "preprocessing done\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|███████████████████████████████████████████████████████| 21/21 [1:01:41<00:00, 176.26s/ep, tloss=35.4, vloss=1.14]\n"
     ]
    }
   ],
   "source": [
    "NET_NAME = \"trump.torch.net\"\n",
    "\n",
    "def train(train_file: PosixPath, val_file: PosixPath, \n",
    "          epochs: int, word_len: int, batch_size: int, lstm_cells: int, lr: int, use_wordstarts=True, net=None):\n",
    "    # data loaders\n",
    "    transcoder = Transcoder(train_file)\n",
    "    n_signs = len(transcoder.decoder)\n",
    "    train_dataset = TextDS(train_file, word_len, transcoder, use_wordstarts=use_wordstarts)\n",
    "    val_dataset   = TextDS(val_file,   word_len, transcoder, use_wordstarts=use_wordstarts)\n",
    "    # use 1 worker only, because windows can't pickle lambdas/gives a broken pipe\n",
    "    train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle=True)\n",
    "    val_loader   = torch.utils.data.DataLoader(val_dataset,   batch_size=batch_size, shuffle=True)\n",
    "    print(\"preprocessing done\")\n",
    "    \n",
    "    # init net parameters\n",
    "    if not net:\n",
    "        net = Model(n_cells = lstm_cells, n_output=n_signs).to(device)\n",
    "    criterion = nn.BCELoss()\n",
    "    optimizer = optim.Adam(net.parameters(), lr=lr)\n",
    "    progress = tqdm(range(epochs), unit=\"ep\")\n",
    "    history = []\n",
    "    \n",
    "    # training loop, visualized with tqdm\n",
    "    for epoch in progress:  # epoch loop\n",
    "        # training loop\n",
    "        tloss, vloss = 0, 0\n",
    "\n",
    "        for inputs, labels in train_loader:\n",
    "            optimizer.zero_grad()\n",
    "            outputs = net(inputs)\n",
    "            labels = nn.functional.one_hot(labels, n_signs).float()\n",
    "            loss = criterion(outputs, labels)\n",
    "            tloss += loss.item()  # type conversion + unpacking from tensor to float\n",
    "            loss.backward()\n",
    "            optimizer.step()\n",
    "            \n",
    "        # validation\n",
    "        for inputs, labels in val_loader:\n",
    "            outputs = net(inputs)\n",
    "            labels = nn.functional.one_hot(labels, n_signs).float()\n",
    "            loss = criterion(outputs, labels)\n",
    "            vloss += loss.item()\n",
    "        \n",
    "        progress.set_postfix(tloss=tloss, vloss=vloss, refresh=False)\n",
    "        history.append((tloss/len(train_loader), vloss/len(val_loader)))  # length-normalized loss\n",
    "    return net, transcoder, np.array(history).T\n",
    "\n",
    "# pretrain with 100 epochs, use_wordstarts=True, word_len of 25 (less data, 30s /ep)\n",
    "net, transcoder, history0 = train(\"trump_train.txt\", \"trump_val.txt\", epochs=100, word_len=25, batch_size=1024, lstm_cells=100, lr=0.01, use_wordstarts=True)\n",
    "# torch.save(net, NET_NAME)\n",
    "# finetune with 21 epochs, use_wordstarts=False, word_len of 50 (more data, 300s/ep)\n",
    "net, transcoder, history1 = train(\"trump_train.txt\", \"trump_val.txt\", epochs=21 , word_len=50, batch_size=1024, lstm_cells=100, lr=0.001, use_wordstarts=False, net=net)\n",
    "# torch.save(net, NET_NAME+\".finetuned\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2 style=\"color:rgb(0,120,170)\">Visualize the training progress</h2>\n",
    "\n",
    "Now we visualize the training progress we recorded before.\n",
    "\n",
    "The fine tuning loss is higher in the fine tuning, as we have more and new data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def visualize(history, title):\n",
    "    plt.rcParams[\"figure.figsize\"] = (10,8)\n",
    "    plt.plot(history[0])\n",
    "    plt.plot(history[1])\n",
    "    plt.xlabel(\"epoch\")\n",
    "    plt.ylabel(\"loss\")\n",
    "    plt.legend([\"train\", \"test\"])\n",
    "    plt.title(title)\n",
    "    plt.show()\n",
    "    \n",
    "visualize(history0, \"pretrain with 100 epochs, use_wordstarts=True, word_len of 25 (less data, 30s /ep)\")\n",
    "visualize(history1, \"finetune with 21 epochs, use_wordstarts=False, word_len of 50 (more data, 300s/ep)\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Training + test loss over time"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2 style=\"color:rgb(0,120,170)\">Text generation</h2>\n",
    "\n",
    "To predict text, we want the option to deterministically predict the most probable text (greedily always take the most probable next char) or to non-deterministically sample one of the 5 most-probable characters with their respective chance given by the output of the network."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "let me say \n",
      "in people. they want to be tremendous to to tell to be a tride and the corerasion. what we're going to started all over\n",
      "which it. and it stive out people. and who going to be the litered fumst of these people in the place. we always that\n",
      "incredible. they dide they don't have to have all over that they want to tell you it was in serising the chapace than\n",
      "and i would and the word in plant where the company. they'ls so here walking that a lot the people and we sort out and\n",
      "i'm going to sill think then i would actueling a trayente anything on sorrars. we have and all of these some of thes\n",
      "\n",
      "i think that \n",
      "is samizations. that's and it's not all terrination. you know. they differen about a plentiant on trus the wasn't say\n",
      "than they didn't want tellicion. you said, and we said, \"you knows than we do what when i think of that's alongs. we\n",
      "the belistor. i talk an impigit out a great. and i'm the country, any doesn't be all, i do the wife are going to stop\n",
      "are the country. they went to here to an sority. i have nothing. they're seweence. they look a long. we have to said\n",
      "with a probably inceront it. they have now we're not guy is saying, i did out, and when we're not going tremendout,\n",
      "\n",
      "we should \n",
      "heveon, same that's gold saugh these. it'ss in treating and it so is a lives that around it. as the poll, they do this\n",
      "get our the way, you wad told to strong porting, i shir, to a little second the mole. we're going to be treen, it was\n",
      "have to think which tremendous a great that's not is and seed our country week on man, if they're. i soow and so hands\n",
      "to tell you this is golded it. we're says. we are going to get to belleve of them, they said, \"he're going to get of\n",
      "have all these pall and i have so the begn this against that always with it. and they're going to be the sappe and\n",
      "\n",
      "i am the \n",
      "dowem of people well, we're going to build that's an some people is what's great is a lot of there were to this is been\n",
      "wall that and well, i won and they are goes withinkess they are to the musse people. and i'm a great cimple is a lot of\n",
      "big a livered a thinkerint. they did is an incrediblis are good telarney that. the way well. and there's going to be\n",
      "world the what the reason are great to there a protable than we're deal office seed the mounden this and so why\n",
      "day and somebody a trump: if you started in all out a disaster, they do i have stupin, a lot of a the some of at, and\n",
      "\n",
      "the best \n",
      "parts. if you know the pralticles and he'll do i've end have say i don't had always any stifiin, and i will have a\n",
      "a loous shourd have bades that all is back. we said, the losmerts there. and and they do alebs policyed to say, \"they\n",
      "trump, and if that's so missive that him to that is strengent into a lond and we're story was to then they went by the\n",
      "of tough or a lot. you know, it's going to have been a then he was a moneys a disaver facter. the chade. i would amout\n",
      "or any well. they said they well are. i have the weaks intermactule on my was a trade been so. so the countrisians a\n",
      "\n",
      "we should \n",
      "americuls and i says that thing. i want to say, \"we had stung to say, and are a traded. i was going to say that a\n",
      "guy there. we saup i do we have been a thing tite i want. and the was out in said we have a deals in importes indo the\n",
      "have they want to terrorisial of the corally was incredible stouble. you know what they do into them was an all white\n",
      "be the meally wis and then it, they'll have back we're going to go bado'l saying in two treats we can't hast, and\n",
      "go accome that some of state a tal the president they want to see things. so we're going to do a country. we're a\n"
     ]
    }
   ],
   "source": [
    "def generate(net: Model, inp: list, out_len: int, nr_pick=1, with_chances=True) -> list:\n",
    "    inp = tensor([inp]).long().to(device).requires_grad_(False)  # wrap in list for batches\n",
    "    for _ in range(out_len):\n",
    "        out = net.forward(inp)[0]\n",
    "        out = out.detach().cpu()  # transfer from gpu to cpu, so we can use it\n",
    "        # take the last N options and randomly pick one of them\n",
    "        if with_chances:\n",
    "            best = out.argsort()[-nr_pick:]\n",
    "            chances = out[best].numpy()\n",
    "            nxt = tensor(np.random.choice(best, 1, p=chances/chances.sum()), requires_grad=False).to(device)\n",
    "        else:\n",
    "            nxt = out.argsort()[-nr_pick:]\n",
    "            nxt = nxt[torch.randint(nr_pick,(1,))]\n",
    "        \n",
    "        nxt = torch.as_tensor(nxt).unsqueeze(0) # ..as_tensor\n",
    "        inp = torch.cat((inp, nxt), 1)\n",
    "    return inp[0, -out_len:].tolist()\n",
    "\n",
    "for start_text in [\"let me say \", \"i think that \", \"we should \", \"i am the \", \"the best \", \"we should \"]:\n",
    "    print(\"\\n\"+start_text)\n",
    "    for i in range(5):\n",
    "        start_text_enc = transcoder.encode(start_text)\n",
    "        gen_text_enc = generate(net, start_text_enc, 120, 5, with_chances=True)\n",
    "        gen_text = transcoder.decode(gen_text_enc)\n",
    "        print(gen_text.replace(\"\\n\", \" \")[:gen_text.rindex(\" \")]) # for nicer printing, cut half-words"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "Output after pretraining without finetuning:\n",
    "\n",
    "let me say ...\n",
    "\"well, our military. you know, you heart the gristh said us. doing this aum and they were time the w\n",
    "\"ohay?\" and with you can't even went it, \"ob approstar economize in mexico, francitional save people\n",
    "what he's going to get out the cond of the sublock. he's some paperselder is greatents. but they get\n",
    "and i called a poing of the stands, as going to be able many people hosore socces four days, we're g\n",
    "what how militarule are in the way it wasn't. it's not the rupes crittor. nobody that time first lik\n",
    "\n",
    "i think that ...\n",
    "is me the muttle waiting they suppost business. i shouse heard american enelf companies that we here\n",
    "we don't want to haptes folks. i tell you this time you have been i have the can because i don't hav\n",
    "i think we need teng it work beroused, sect, a deal. i don't know how we will me be out we doing tim\n",
    "i was and a from they did them to have the crowd. and we have to be talking about. she shising in a \n",
    "we have been taken the burded all of a person... i'm going to make the statial. and he said terricac\n",
    "\n",
    "we should ...\n",
    "me an expreciate lough and thosp, movel but you want to have to build and we have a problem. these p\n",
    "think they're a bad. you want to have to do with a presian. i said, \"we have nice person about must \n",
    "me be gotting. i don't know.\" income in cledes. we all in they've see workde sobign tso but i well n\n",
    "are getting from chrowther are me so with way. that was so great foreate. in the roster. they're all\n",
    "are a lot of this\" so we have that is going to be taking away from anymore. well, i'll mek out they'\n",
    "\n",
    "i am the ...\n",
    "smartes...that we need to do it. after and they don't louble trade islam. we doing by a firuca. we w\n",
    "money and we were the should the corce think is spert people to they we'rred to say it. i don't know\n",
    "mentide, i was grist. we will have a little sturized have a parth. if you look at mexico in allowed \n",
    "strongle that. i was again. but we have this and all of the wall imazing with the cridge he'd beated\n",
    "book at be a good person for the stefuct press that i was going to be pay that the statuty with the \n",
    "\n",
    "the best ...\n",
    "same defactaces as something. you know, he said \"we're going to have no goomedss are we shicldy peop\n",
    "is a lotties are they tell the world. and he comes talking. nobody ever that is that we do i have th\n",
    "i think the other rubio? cordome has been the bract staying, hosest for a long. \"dand folks. we can'\n",
    "is it. i didn't even are a lot of cursed in trump complet and they're all there. they want to be str\n",
    "ince. they're geing to be story i didn't have no honor that he was askid cristly that's all of the w\n",
    "\n",
    "we should ...\n",
    "are in orrucs of the stuff with me so well in a folks, you know they'll do this and the press here.\"\n",
    "think folks wife i don't know how they've say, \"well, they can herr. well, we're going to see this s\n",
    "make a bedicip wack. and he was a big, with mentiont. and if i dicn, they did over there was specian\n",
    "carraighted that i don't want you tell you this countries, that's here and they'd leaders are make a\n",
    "think of a called so many thought bad docticasts arane was great thought the statistic, i heard in t"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}